Analysis
Kinetix AI -- not to be confused with the unrelated metaverse-content startup of the same name -- has raised $75 million in an Angel+ round led by Temasek's Vertex Ventures, with backing from Fangguang Capital and Vanshi Capital, VentureBurn reported. Founded in September 2025, the company is barely a year old, and the round is unusually large for a startup at that stage -- a pattern that has become common in embodied AI, where hardware and data costs front-load spending well before any product ships.
Kinetix AI describes itself as a full-stack embodied-intelligence company, building everything from a native embodied large language model to a data-collection headset used to capture human motion for training, an AI infrastructure platform spanning data collection through simulation and physical deployment, and its flagship product: KAI, a full-size humanoid robot with 115 degrees of freedom. The founding team draws on veterans of Huawei's autonomous-driving division alongside researchers from the University of Hong Kong -- a lineage that mirrors how much of China's humanoid-robotics wave has been staffed by engineers who cut their teeth on self-driving perception and control systems before pivoting to legs and arms.
The strategic bet embedded in the round: Kinetix AI is explicitly targeting full-size, general-purpose humanoids rather than cheaper wheeled or tabletop robots that dominate near-term commercial deployment. That's the same wager Figure AI, Unitree, and Agility Robotics are making in the US and elsewhere in China -- betting that general-purpose form factors win the category even if task-specific robots reach revenue faster.
The new capital goes toward what the company calls its "data-model-hardware" iteration loop -- collecting motion data, retraining the model, testing in simulation, deploying to physical hardware, and repeating. That loop is the actual bottleneck in humanoid robotics right now, more than any single component: training data for real-world manipulation is scarce, expensive to collect, and doesn't transfer cleanly between robot bodies, which is why an unusually large seed-stage round buys runway rather than a finished product.
